MegaTroll222

Omni Ugc Ad Factory — Development skill for Claude Code

Development community

Turn a product + an actor reference photo into a seamless ~20s UGC talking-head ad — Claude Code skill with machine-enforced prompt gates, on the MaxFusion MCP.

How to install Omni Ugc Ad Factory

This entry records only its repository, not the path inside it, so there is no exact command to give. Open MegaTroll222/OMNI-UGC-AD-FACTORY and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Omni Ugc Ad Factory does

Turn a product + an actor reference photo into a seamless ~20s UGC talking-head ad — Claude Code skill with machine-enforced prompt gates, on the MaxFusion MCP.

Alternatives in Development

Codexmcp — Enable seamless collaboration between Claude Code and Codex, transforming from a single agent to multiple agen 2k ★
Vox Collage Broll — Turn one spoken line into a paper-collage explainer video — Claude Code skill + MaxFusion MCP 198 ★
Technique Combine People — Combine a user's photo with another person (real or fictional) into a single composite image: $ARGUMENTS 178 ★

README

Omni UGC Ad Factory

A Claude Code skill that turns a product and an actor reference photo into a finished **~20-second UGC talking-head ad**: two chained 10-second Gemini Omni clips that play as one unbroken take, shipped as a single seamless 9:16 MP4. Built on the [MaxFusion](https://maxfusion.ai) MCP.

Real outputs of this pipeline (click a preview for the full MP4 with sound):


![gym](examples/final_916_gym.gif)	![room](examples/final_916_room.gif)	![room alt](examples/final_916_room_alt.gif)

Why it exists

Anyone can prompt a talking head. The hard parts are everything that makes viewers believe it: an actor that looks captured rather than rendered, a script that sounds like a person and not an ad, a voice that isn't monotone, micro-behaviors (uneven pauses, a gaze drop, a nose exhale) that land on the right beats, and two clips that join without a visible seam.

This skill encodes all of that as **machine-enforced gates**, not prose advice:

casting.json  ->  gate_casting.py  ->  approved_casting.txt  ->  gpt-image-2
clip1.json    ->  gate_omni.py     ->  approved_clip1.txt    ->  gemini-omni-flash
clip2.json    ->  gate_omni.py     ->  approved_clip2.txt    ->  gemini-omni-flash

The agent never hand-writes a generation prompt. It writes structured JSON; the gate validates it against data-file glossaries (age-correct skin physiology, a 32-entry micro-behavior glossary keyed to internal states, a six-category voice spec, a camera color-grading vocabulary) and *assembles* the prompt itself on PASS. A prompt that skipped its gate does not exist.

The chaining trick

Omni holds its reference image for the first frames of a clip before dissolving into the generated scene. Seed clip 2 with clip 1's trimmed final frame and that quirk becomes an invisible join — measured seams of ~3/255 against a hard gate of 5/255. `scr